running a woodchipper through a woodchipper doesn’t help you understand the woodchipper
it is really funny though. i’ll give them that
one of sigmund freud’s pupils once said “sometimes a cigarette is just a cigarette”
Do you think you could build a woodchipper without the understanding of the woodchipper
we would go outside to record. I chose Not To though and someone else did it for me. In getting someone else to do it for me I have befriended the most tolerable person in my class because hes just a Normal Guy. he does normal guy things.
when in high school i delayed the hell out of my music tech assignments because they required us to record our own voice. i cant do voice recordings even worse
don’t humans do this too
no i was mostly being pithy, i understand how neural nets can develop beyond their creators’ understanding, it’s just, like
“we literally identified the specific neurons involved in producing LLM outputs [in some circumstances, because this is in fact hard to do]”
“you used an AI to help you rather than ???somehow having a human personally analyze over a billion parameters?? so clearly you aren’t really trying to understand it. I am very smart”
put people through woodchippers?
1 Like
Yeh
we run humans through woodchippers?
1 Like
i don’t do it to understand them
1 Like
of course if you run a model through a similar model it’s never going to output “the entire foundation of our field is flawed”, because if the entire foundation of the field is flawed then it wouldn’t fucking know that
fuck you fuck you fuck you
1 Like
too slow
Idk much about the topic but human brains use human brains to understand human brains so I don’t see the problem using ai to help better understand ai’s
oh gods are we getting close to P vs. NP posting because i feel we’re getting close here
i like when ai is forced to learn off other ai because it slowly but very surely breaks apart until they become totally incoherent
1 Like
for reference i also think that psychology is a fundamentally flawed field that must be utterly reconstructed from first principles
1 Like
like idk, to me all of this AI talk is done with bad perspective
The purpose of the tech is to be able to handle things that it would take millions of humans to do as a conglomerate, instead we just built that conglomerate in a lab.
Like why would we expect a single human to understand the conglomerate made for millions